"This paper evaluates empirically the performance of six structural credit risk models by comparing the probabilities of default (PDs) they deliver to ex post default rates. In contrast to previous studies pursuing similar objectives, the paper employs firm-level data and finds that theory-based PDs tend to match closely the actual level of credit risk and to account for its time path. At the same time, nonmodelled macro variables from the financial and real sides of the economy help to substantially improve the forecasts of default rates. The finding suggests that theory-based PDs fail to fully reflect the dependence of credit risk on the business and credit cycles. Most of the upbeat conclusions regarding the performance of the PDs are due to models with endogenous default. For their part, frameworks that assume exogenous default tend to underpredict credit risk. Three borrower characteristics influence materially the predictions of the models: the leverage ratio; the default recovery rate; and the risk-free rate of return"--Bank for International Settlements web site.